Edition: April 21, 2026

Meta plans to record its employees' keystrokes and mouse movements to train its AI models, sparking concerns over privacy implications. The data will be used to help models learn how people interact with computers. The move highlights the lengths tech companies are going to find new sources of training data.

A group of unauthorized users has gained access to Anthropic's exclusive cybersecurity tool Mythos, sparking concerns about potential misuse. The group, which is part of a private online forum, obtained access through a third-party vendor. Anthropic is investigating the incident, but so far, no evidence of impact on its systems has been found.

SpaceX has partnered with coding platform Cursor to develop a next-generation AI, with an option to acquire the startup for $60 billion later this year. The deal aims to combine Cursor's expertise with SpaceX's Colossus supercomputer, enhancing its AI capabilities. This move is seen as a strategic step ahead of SpaceX's anticipated public offering.

SpaceX has announced a potential deal to acquire automated programming platform Cursor for $60 billion, which could help its xAI tools compete with market leaders like Anthropic. The deal includes a $10 billion fee if the acquisition doesn't go through. This move comes as major tech companies like Google and OpenAI are also focusing on improving their AI coding capabilities.

As Apple prepares for a leadership transition, the company has an opportunity to rediscover its humanity and return to its roots, focusing on innovation and customer needs rather than profit. The incoming CEO, John Ternus, can reorient the company and make it more customer-centric. This shift could help Apple regain its reputation and stay relevant in a rapidly changing tech industry.

A recent MIT study reveals that inversions, where warm air settles over cooler air, can lead to hotter, muggier, and stormier summers in midlatitude regions. This phenomenon is expected to worsen with global warming, potentially turning the US Midwest and eastern Asia into new hotspots for humid heat waves. The study provides a key understanding of the limits for humid heat and severe convection in these regions.

MIT is witnessing a pervasive integration of AI across various departments and labs, transforming research in fields like energy, aerospace, and life sciences. Researchers are leveraging AI tools to develop innovative solutions, such as digital twins and optimized material compositions. This widespread adoption of AI is revolutionizing the way research is conducted at MIT.

Researchers propose using transcranial focused ultrasound to study consciousness, allowing for noninvasive exploration of brain activity. This technology can reach deeper into the brain with greater resolution than existing methods. It may help test competing theories of consciousness, shedding light on the neural basis of subjective experience.

David Huang, a clinician-scientist, recalls the breakthrough of optical coherence tomography (OCT), a noninvasive eye imaging technology he invented that is now used in 40 million procedures per year. OCT produces detailed images of biological tissues such as the retina and coronary arteries. Huang's discovery has earned him numerous awards, including the Lasker Award and the National Medals of Technology and Innovation.

MIT engineers have developed a method to 3D-print construction-grade beams and trusses using recycled plastic, offering a sustainable alternative to traditional wood-based framing. The printed structures are lighter, more durable, and can bear loads exceeding key building standards. This innovation could help address both plastic pollution and the need for sustainable housing solutions.

Researchers at MIT have discovered that a natural protein called intelectin-2 can protect the GI tract from infection by binding to sugars on bacterial membranes and strengthening the intestine's protective lining. This protein could potentially be adapted as an antimicrobial agent to combat antibiotic-resistant bacteria. It may also help treat disorders such as inflammatory bowel disease.

Researchers have found evidence that early life forms may have evolved to use oxygen hundreds of millions of years before the Great Oxidation Event, challenging previous beliefs. This discovery was made by tracing the origins of an oxygen-enabling enzyme to 3.2-2.8 billion years ago. The findings may explain why oxygen didn't build up in the atmosphere until later.

Researchers at MIT have developed an analog computing method that utilizes waste heat from electronic devices to perform data processing, achieving over 99% accuracy in simple matrix vector multiplication. This innovative approach could potentially reduce energy consumption and eliminate the need for extra temperature sensors. The technique may have immediate applications in detecting heat sources and measuring temperature changes in electronics.

Apple's incoming CEO John Ternus will inherit a multitude of challenges, including ongoing antitrust battles, tensions with governments over encryption, and a delicate balancing act with China. Ternus takes the reins from Tim Cook, who has navigated these issues during his 15-year tenure. The new CEO will need to address these complex problems while driving the company's growth and innovation.

Meta will track US employees' mouse movements, clicks, and keystrokes to generate training data for future AI agents, using a new software called Model Capability Initiative. The data will be collected from specific work-related apps and websites, along with periodic screenshots. This initiative aims to improve Meta's AI models through employee interactions.

NeoCognition, a startup developing self-learning AI agents, has emerged from stealth with $40 million in seed funding to build agents that learn like humans, aiming to make AI more reliable and efficient. The company's technology focuses on creating agents that can specialize in any domain, similar to human learning. NeoCognition plans to sell its agent systems primarily to enterprises, including SaaS companies, to enhance their product offerings.

OpenAI's updated image generator, ChatGPT Images 2.0, can now pull information from the web to create more sophisticated images from a single prompt. This is made possible by OpenAI's new GPT Image 2 model, which is available to certain subscribers. The update improves the generator's ability to follow instructions and preserve details.

OpenAI has launched ChatGPT Images 2.0, a new image generation AI model that can generate multiple images from a single prompt and output text in non-English languages. The model can tap into ChatGPT's reasoning capabilities and has a more recent knowledge cutoff date, allowing for more thorough and accurate generations. This update is available globally for ChatGPT and Codex users, with a more powerful version available for paying subscribers.

OpenAI's new ChatGPT Images 2.0 model has significantly improved its ability to generate realistic text within images, allowing for the creation of complex visuals like menus and comic strips. The model's capabilities include searching the web, making multiple images from one prompt, and double-checking its creations. This advancement brings image generation closer to human-like quality, with outputs available up to 2K resolution.

Mozilla used Anthropic's AI tool, Mythos, to identify and fix 151 bugs in its Firefox browser, highlighting the potential impact of AI on cybersecurity. The company believes that emerging AI capabilities will create a new standard for software security, requiring a coordinated effort to find and fix latent vulnerabilities. This development comes as AI models like Mythos are becoming increasingly powerful in uncovering software flaws.

Framework has introduced its first eGPUs, allowing users to turn their laptop into a desktop PC by plugging in a desktop graphics card or other PCIe cards. The OCuLink Dev Kit uses the OCuLink standard to transmit data between the CPU and external GPU. This innovation enables laptops to have more power than most laptops ever dream of having.

As Tim Cook's tenure as Apple CEO comes to a close, his legacy is marked by the company's significant growth in its services business, including iCloud, Apple Music, and Apple TV+, which has become a major revenue stream. The services business reached an all-time revenue record of $30 billion in the last quarter, with over $109 billion generated in fiscal year 2025. New CEO John Ternus will face the challenge of extending Apple's platform into the generative AI era.

YouTube is expanding its AI deepfake monitoring feature to include celebrities, allowing them to find and request removal of AI-generated content featuring their likeness. The feature uses likeness detection to flag deepfakes for enrolled public figures. Removal requests will be evaluated against YouTube's privacy policy.

As AI agents become more prevalent in organizations, companies are at risk of introducing new attack surfaces if they don't implement robust governance and security measures. A recent Deloitte report found that nearly 74% of companies plan to deploy agentic AI within two years, but only 21% have a mature model for governance. Implementing a control plane that governs and secures AI agents is critical to mitigating these risks.

Clarifai has deleted 3 million photos provided by OkCupid to train its facial recognition AI, following an FTC investigation that found the data was shared without user consent. The incident occurred in 2014, but was only recently settled, with OkCupid and its parent company Match Group prohibited from misrepresenting their data collection practices. The deletion of the data and models trained on it marks a conclusion to the case.

NeuBird.ai is developing agentic AI systems to automate incident response, using AI minions to handle investigations behind the scenes, aiming to reduce human dependency on observability. The company plans to integrate with other parts of the incident response chain, closing the SRE/DevOps loop. This technology has the potential to revolutionize the field of site reliability engineering and DevOps.

Latitude, the maker of AI Dungeon, has unveiled Voyage, a platform for creating AI-powered RPGs with unscripted interactions and customizable game worlds. The platform allows users to design their own gaming worlds with the help of AI and share them with others to play. Voyage offers a range of experiences across different genres, with character progression and unexpected interactions driven by AI.

YouTube is expanding its AI likeness detection technology to protect celebrities from deepfakes, allowing them to request removal of videos that use their simulated faces without permission. The technology, similar to YouTube's Content ID system, scans for visual matches of an enrolled participant's face. It will also support audio detection in the future.

Google DeepMind has partnered with industry leaders Accenture, Bain & Company, BCG, Deloitte, and McKinsey to accelerate AI transformation and bring frontier AI to organizations worldwide, aiming to drive global economic growth and solve complex challenges. The partnership will focus on critical enterprise needs in sectors like finance and manufacturing. This collaboration is expected to help businesses harness the potential of AI responsibly and at scale.

Apple's new CEO, John Ternus, faces a significant challenge in addressing the company's lag in AI development, despite being a 25-year veteran and leading hardware engineering for various Apple products. Ternus will succeed Tim Cook as CEO on September 1st. The company's lack of AI announcements has raised concerns in the past.

Vercel, a tech company, was breached by AI-assisted intruders who used OAuth abuse and a pilfered employee account to gain access to the company's systems. The attackers, suspected to be aided by AI, moved with 'surprising velocity' and a deep understanding of Vercel's infrastructure. Stolen data is being sold for $2 million on BreachForums.

A medical student from India created an AI-generated 'MAGA girl' to scam conservative men online, making thousands of dollars by posting provocative content and selling merchandise. The scammer used a chatbot to refine his approach, targeting a niche audience with high disposable income. This is part of a larger trend of AI-generated influencers infiltrating social media.

Researchers have introduced QIMMA, a quality-first Arabic LLM leaderboard that validates benchmarks before evaluating models, ensuring reported scores reflect genuine Arabic language capability. QIMMA consolidates 109 subsets from 14 source benchmarks into a unified evaluation suite of over 52,000 samples. The platform aims to address the fragmented and unvalidated Arabic NLP evaluation landscape.

A research team led by Jochen Knies has embarked on an expedition to the North Pole to study the region's past and determine if it had been ice-free in recent Earth history. The team collected sediment cores from the Arctic seafloor to analyze climate archives and understand the impact of sea ice loss on the marine ecosystem. The findings could provide insights into the future of the Arctic and global climate.

Siemens has introduced the Eigen Engineering Agent, an AI system designed to automate engineering tasks in operational environments, increasing efficiency and accuracy. The system can execute tasks two to five times faster than manual workflows and is integrated into Siemens' Totally Integrated Automation Engineering platform. This development aims to address the global shortfall of manufacturing workers and skills gap in industrial environments.

Researchers have developed Nemotron-Personas-Korea, a dataset of 6 million synthetic personas grounded in official Korean statistics, to help AI agents better understand and serve Korean users. This dataset provides a more accurate and culturally sensitive context for AI interactions. The personas can be used to deploy a Korean AI agent in about 20 minutes using hosted APIs.

The emergence of AI models like Mythos is revolutionizing cybersecurity by rapidly finding and patching software vulnerabilities, and openness in code and tooling can be a structural advantage in this field. As autonomous systems proliferate, open ecosystems can help level the playing field and distribute knowledge and action across communities. This shift has significant implications for the future of cybersecurity and the role of AI within it.

Snowflake is expanding its Snowflake Intelligence and Cortex Code platforms to bring AI capabilities to both non-technical and technical users, with new features including increased integrations, automation, and simplified workflow building. The updates aim to make AI more accessible and personalized for users. Snowflake Intelligence and Cortex Code are designed to work with various third-party software and data sources.

GRAI, a new music startup, believes AI can make music more social by allowing users to remix and interact with tracks, rather than replacing artists. The company aims to give artists control over how their music is used and plans to capitalize on the power of AI to transform music engagement. With $9 million in seed funding, GRAI is developing apps that enable users to play with tracks and create new royalty payments for artists.

MIT Technology Review's EmTech AI conference unveiled a list of 10 key technologies and trends in AI to watch in 2026. The list, presented by executive editors Amy Nordrum and Niall Firth, highlights emerging ideas and movements in the field. The session was hosted by AI reporter Grace Huckins and is available to stream online.

A growing backlash against AI is emerging in the US, with many Americans expressing concerns over the technology's impact on economic stability and public safety. Despite this, AI has yet to become a major focus of election campaigns. Experts say that could change as more people demand regulation of the technology.

OpenAI CEO Sam Altman criticizes Anthropic's cybersecurity model, Mythos, as using 'fear-based marketing' to make the product sound more impressive. Altman implies that this tactic is used to keep AI in the hands of a select few. The criticism comes amid ongoing competition between OpenAI and Anthropic.

Bond, a new social media platform, aims to help users overcome their screen addiction by using AI to generate personalized recommendations for real-world experiences. The platform allows users to post about their experiences and stores them in a private archive, which is then used to provide tailored suggestions. By doing so, Bond encourages users to engage in offline activities rather than mindless scrolling.

Google Ads is introducing three new safety features in Ads Advisor to make the platform safer and faster, including proactive troubleshooting, 24/7 security monitoring, and instant certifications. These features aim to reduce the time spent managing campaigns and increase business growth. The updates utilize AI to flag policy violations, monitor account security, and automate certifications.

Human noise is negatively impacting animals, with technological solutions being explored to mitigate the issue, while Los Angeles is set to launch a new subway segment connecting downtown to the Pacific Ocean. Additionally, Apple's CEO Tim Cook is stepping down, and Anthropic has signed a major compute deal with Amazon.

Yelp is upgrading its AI chatbot to become a digital concierge, allowing users to get things done more efficiently. The chatbot will be able to answer questions, make recommendations, and handle bookings in a single conversation. This move is part of a broader industry push to make AI more relevant and useful to consumers.

Starbucks' new ChatGPT integration has been met with frustration, as a simple order turned into a complicated process. The author, who typically orders with ease, found the experience to be a 'coffee nightmare'. The integration, which launched last week, allows users to order by typing '@Starbucks' plus their order in ChatGPT.

The MIT community has released several new books covering a range of topics including technology, economics, and social justice. Titles such as 'Priority Technologies' and 'The Art of Monetary Policy' offer insights into the future of work and global economics. These books are now available from various publishers.